
Bode Plot Gain and Phase: Where Feedback Can Go Wrong
A Bode plot shows how a linear time-invariant (LTI) system changes a sinusoidal signal’s amplitude and phase as frequency varies. Misread its gain crossover in a feedback amplifier, and a design you expect to settle can instead oscillate. Reading the magnitude and phase panels together reveals how feedback can approach that boundary.
In This Article
- Read the Two Panels Together
- Combine Simple Transfer-Function Factors
- Work Through a First-Order Low-Pass Response
- Account for Resonance and Time Delay
- Measure Margins at the Correct Crossovers
- Draw the Response From Your Model
- Key Takeaways
- Frequently Asked Questions
- Check the Response Before Judging Stability
Read the Two Panels Together
An amplifier’s output can swing repeatedly instead of settling because the returning feedback reinforces a disturbance. A magnitude curve alone cannot explain that failure: you also need the phase at the same frequency. Bode plots put both on a shared logarithmic frequency axis.
The magnitude panel usually shows (20\log_{10}|G(j\omega)|) in decibels; the phase panel shows (\arg G(j\omega)) in degrees. Together they describe the amplitude ratio and phase shift between a sinusoidal input and its steady-state output. At 0 dB, amplitudes are equal; positive values mean amplification and negative values mean attenuation. Phase identifies lead or lag.
The frequency axis may use radians per second or hertz, related by (\omega=2\pi f). A decade means a tenfold increase, such as 1 to 10 rad/s. That spacing lets you follow changes across a wide frequency range without losing the low-frequency detail.
Combine Simple Transfer-Function Factors
Factor the transfer function before sketching it: multiplication becomes addition of magnitudes in decibels and phases in degrees. The useful contrast is between factors that increase the slope and those that decrease it.
A first-order left-half-plane zero turns the asymptotic slope upward by 20 dB/decade above its corner; a pole (1/(1+s/\omega_p)) turns it downward by the same amount above (\omega_p). Their phases move in opposite directions: the pole goes from approximately 0° toward −90°, passing −45° at its corner, while the zero contributes the opposite phase. The actual curves bend through these straight-line approximations.
At the origin, the contrast is simpler: a differentiator (s) contributes +20 dB/decade and +90°, while an integrator (1/s) contributes −20 dB/decade and −90°. A positive constant (K) shifts magnitude by (20\log_{10}K) without changing phase.
Work Through a First-Order Low-Pass Response
For (G(s)=1/(1+s)), the corner is 1 rad/s. Substituting (s=j\omega) gives
[ |G(j\omega)|=\frac{1}{\sqrt{1+\omega^2}}, \qquad \arg G(j\omega)=-\tan^{-1}(\omega). ]
At 1 rad/s, the response is −3.01 dB and −45°. At 10 rad/s, it is −20.04 dB and −84.3°. These points locate the rounding that a straight-line sketch misses.
The usual −3 dB bandwidth is relative to passband gain. Here the passband gain is unity, giving a bandwidth of 1 rad/s, approximately 0.159 Hz. Raising the response with a positive constant leaves that corner unchanged. An absolute 0 dB crossing measures unity-gain crossover, not passband-relative bandwidth. Depending on the constant gain, that crossing can fall below or above the bandwidth, or may not exist.
Account for Resonance and Time Delay
A second-order response can overshoot after a step and ring as it settles. For the standard low-pass factor
[ G_2(s)=\frac{\omega_n^2}{s^2+2\zeta\omega_n s+\omega_n^2}, ]
(\omega_n) is the natural frequency and (\zeta) the damping ratio. Below critical damping, with (0<\zeta<1), an oscilloscope shows decaying excursions around the final value. A resonant magnitude peak requires the narrower range (0<\zeta<1/\sqrt{2}). The pair eventually contributes −40 dB/decade and −180° phase.
A delay can threaten feedback stability without any magnitude peak. The factor (e^{-sT}) leaves magnitude unchanged but adds (-\omega T) radians, or (-\omega T,180/\pi) degrees. Its phase contribution curves on a logarithmic frequency axis, consuming margin while the magnitude panel looks untouched.
Measure Margins at the Correct Crossovers
Use the loop transfer function (L(s)) for standard negative feedback, including the controller and relevant sensor dynamics or delays, not automatically the closed-loop input-output response. At gain crossover, (\omega_{gc}), the loop magnitude is unity, or 0 dB. Before using the following formula, express the phase there in the interval [−360°, 0°] by adding or subtracting multiples of 360°:
[ \mathrm{PM}=180^\circ+\arg L(j\omega_{gc}). ]
A phase of −135° gives 45° phase margin. Unwrapping removes display jumps but does not by itself set this phase reference. If there are multiple unity-gain crossings, calculate the margin at each one. A positive phase margin alone does not guarantee stability; multiple crossovers, unstable open-loop poles or unusual loop shapes require a complete stability test, often using the Nyquist criterion.
Gain margin reverses the operation. Find the phase crossover, (\omega_{pc}), where phase reaches −180° modulo 360°, then read magnitude:
[ \mathrm{GM}{\mathrm{dB}}=-20\log{10}|L(j\omega_{pc})|. ]
A reading of −10 dB gives 10 dB gain margin. For the simple oscillatory stability boundary, the decisive combination is unity loop magnitude and −180° phase: (L(j\omega)=-1). Then the closed-loop characteristic expression (1+L(j\omega)) is zero, marking the boundary at which sustained oscillation can occur.
Draw the Response From Your Model
Use a hand sketch to understand which factors shape the response; use numerical evaluation to resolve corners and crossovers accurately.
Bode plot methods: choosing how to draw the curves
| Drawing method | Best for | Pros | Cons |
|---|---|---|---|
| Hand-drawn asymptotic Bode plot | Seeing how poles, zeros and gain shape the response | Each first-order zero adds +20 dB/decade after its corner; each first-order pole adds −20 dB/decade | Straight-line magnitude differs by about 3 dB at an isolated first-order corner; narrow resonant peaks need extra calculation |
| Spreadsheet: evaluate G(jω) | Checking a rational transfer function without a control-system library | Use a logarithmic frequency grid; plot 20 log₁₀|G(jω)| and arg G(jω) on separate charts | Requires complex polynomial evaluation, phase unwrapping and enough frequency points to resolve sharp peaks |
| MATLAB: bode(tf(num,den)) | Plotting a numerical transfer function in an existing MATLAB workflow | Automatically generates magnitude in dB and phase in degrees; accepts a specified frequency vector in rad/s | Requires MATLAB and Control System Toolbox licenses; enter polynomial coefficients in descending powers of s |
| Python Control: ct.bode_plot(ct.tf(num,den), dB=True) | Creating reproducible Bode plots with free Python tools | Plots magnitude in dB and phase in degrees; supports transfer-function and state-space models | Requires control and Matplotlib packages; set the frequency range and sample density to capture narrow resonances |
In MATLAB, sys = tf(1,[1 1]); defines the first-order example; bode(sys) plots it. For a loop model named L, margin(L) displays gain and phase margins.
Key Takeaways
- Read the other panel at the same frequency, not at its nearest convenient feature.
- Measure bandwidth relative to the passband, not an arbitrary display level.
- Treat an unchanged magnitude curve cautiously when delay changes.
- Choose numerical evaluation when a sketch cannot resolve a narrow feature.
Frequently Asked Questions
What is a bode plot?
A Bode plot represents sinusoidal steady-state response in separate magnitude and phase panels. If a gain curve looks acceptable but feedback behaves badly, the phase panel may reveal the missing explanation.
What is the difference between a Bode plot and a Nyquist plot?
A Bode plot separates magnitude and phase against frequency; a Nyquist plot traces the response in the complex plane. Nyquist analysis uses encirclements of the critical point −1 to assess stability when individual crossover readings are insufficient.
Why is the magnitude about −3 dB at a first-order corner frequency?
A normalized first-order low-pass pole has an amplitude ratio of (1/\sqrt{2}) at its corner, giving (20\log_{10}(1/\sqrt{2})\approx-3.01) dB. If the plotted corner has a different absolute level, account for the passband gain and other factors before calling it wrong.
Does changing a system’s constant gain change its phase?
A positive constant gain does not change phase. A negative constant introduces a 180° shift, modulo 360°, so a sign error can invalidate your feedback interpretation.
Can a positive phase margin guarantee stability?
No. Phase margin measures the phase separation from −180° at a unity-gain crossover; it is not a complete count of unstable closed-loop poles. If the loop has several unity-gain crossings, a positive margin at one does not describe the others. Unstable open-loop poles also matter: the Nyquist criterion accounts for them together with the loop response’s encirclements of −1. Check every crossover, then verify closed-loop stability using Nyquist analysis or, for a finite-dimensional model, the closed-loop poles. For continuous-time asymptotic stability, all closed-loop poles must lie strictly in the left half-plane; do not let transfer-function cancellations hide unstable internal modes.
Check the Response Before Judging Stability
First establish what your transfer function includes. For filtering, the input-output response answers the bandwidth question. A precise calculation on the wrong transfer function still gives the wrong engineering answer.
Compare the model’s assumptions with the actual feedback path before accepting a stability claim. Software calculates the model you enter; it cannot catch an omitted sensor delay or choose the correct feedback sign for you.
Recommended for you
Inside a Violin Plot: Same Median, Different Shapes
A violin plot can reveal different bill patterns behind the same median. Learn how shape, points, and smoothing affect comparisons. See the restaurant example.
Right vs Left Skewed Histogram: What the Tail Reveals
A right vs left skewed histogram is named for its tail, not its tallest bars. Avoid reversed readings and check averages and binning. Explore the guide.
Frequency Histogram: Every Bar Traced to Its Scores
A frequency histogram shows how common scores are, not how high. Check interval boundaries, trace bars to scores, and avoid misreading the chart. See how.
What Is an Infographic? How Visuals Earn Their Place
What is an infographic? See how an electricity chart and appliance diagram can work together to explain a message, not just decorate it. Explore the basics.
Neural Network Diagram: Labels That Explain the Output
Read a neural network diagram without confusing scores with probabilities. Learn how arrows, layers, and activation labels clarify outputs. See the article.
Pie Chart Examples: The Dollar Totals Behind Each Slice
Use pie chart examples with a $4,000 budget and $100,000 in sales to connect slices to dollars, calculate shares, and compare totals. See the calculations.
Picture Graph Keys: When Four Stars Mean Twenty Books
Read a picture graph without mistaking symbols for counts. See how four stars represent 20 books, use the key, and compare categories in the worked examples.
Flowchart Examples With Labels You Can Make Your Own
Adapt flowchart examples for online orders, leave requests, and login recovery with named actions, decisions, and outcomes. See reusable paths in the article.
